Underwater sonar image segmentation combining pixel-level and region-level information. (May 2022)

• Record Type: Journal Article
• Authors: Chen, Zhe; Wang, Yue; Tian, Wei; Liu, Jutao; Zhou, Ying; Shen, Jie
• Is Part Of: Computers & electrical engineering. Volume 100(2022)
• Publication Date: 2022-05
• Subjects: Side-scan sonar -- Image segmentation -- Pixel-level information -- Region-level information -- Unified Markov random field -- Level set; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2022.107853
• Languages: English
• ISSNs: 0045-7906

Abstract

Abstract: Side-scan sonar (SSS) has been increasingly utilized in underwater exploration, due to the low energy attenuation of the acoustic wave in water. However, SSS imaging is characterized by its high speckle noises, strong intensity bias and low spatial resolution, which pose challenges to accurately delineate objects in underwater sonar images. This issue is handled by our proposed model that segments sonar images via combining pixel-level and region-level information. Pixel-to-pixel and region-to-region neighboring relations are considered in this model. Formally, the unified Markov random field (UMRF) is combined with the level set (LS) to establish a novel energy function for segmenting underwater sonar images. In such a model, the pixel-level information and region-level information are synergistically integrated by UMRF to force LS evolution. Results obtained by our model convergence can accurately segment underwater sonar images into three partitions: the object, shadow and background. From experimental analysis, our proposed model achieves an excellent tradeoff between the segmentation accuracy and noise resistance in underwater sonar image segmentation. The average correctness of our model reaches 0.9431 which has more than 7% improved to the best conventional method.